The Journal of Physical Chemistry Letters · 2022 · 18 citations · 36 references
Natural SciencesComputer ScienceMaterials ScienceComputational Materials ScienceMachine LearningMaterials ChemistryComputational ModelingMaterial ComputationMaterial ModelingForce Field DevelopmentHuman IntelligenceMachine Learning Potential Energy SurfacesMaterials Property PredictionMaterials SimulationWater Force Fields
Currently, computational materials science involves human–computer interaction through coding in software or neural networks. There is still no direct way for human intelligence endorsement. The digitalization of human intelligence should be the ultimate goal for many disciplines. In materials science, human intelligence is still irreplaceable from machine learning techniques, where humans can deal with complex correlations in the real world. We design the framework of Mateverse, a materials science computation platform based on Metaverse, which unifies human intelligence, experiment data, and theoretical simulations. In Mateverse, we intensively study the properties of H2O, including the liquid and solid phases. We show that we can optimize a new water force field (which we name TIP4P-Meta) directly from the interactions between human and visible properties of H2O. This force field is validated to be better than the conventional water model, and new ice polymorphs can be generated. We believe our platform can provide valuable hints in the paradigm upgrade in future computational materials science development.
36
The Graph Neural Network Model
Franco Scarselli, M. Gori, Ah Chung Tsoi et al. · IEEE Transactions on Neural Networks · 2008 · 9.3K citations · Full text
Machine learning and the physical sciences
Giuseppe Carleo, K. Cranmer, Laurent Daudet et al. · Reviews of Modern Physics · 2019 · 2.4K citations · Full text